2 * linux/drivers/block/loop.c
4 * Written by Theodore Ts'o, 3/29/93
6 * Copyright 1993 by Theodore Ts'o. Redistribution of this file is
7 * permitted under the GNU General Public License.
9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
21 * Loadable modules and other fixes by AK, 1998
23 * Make real block number available to downstream transfer functions, enables
24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
25 * Reed H. Petty, rhp@draper.net
27 * Maximum number of loop devices now dynamic via max_loop module parameter.
28 * Russell Kroll <rkroll@exploits.org> 19990701
30 * Maximum number of loop devices when compiled-in now selectable by passing
31 * max_loop=<1-255> to the kernel on boot.
32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
34 * Completely rewrite request handling to be make_request_fn style and
35 * non blocking, pushing work to a helper thread. Lots of fixes from
37 * Jens Axboe <axboe@suse.de>, Nov 2000
39 * Support up to 256 loop devices
40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
42 * Support for falling back on the write file operation when the address space
43 * operations prepare_write and/or commit_write are not available on the
45 * Anton Altaparmakov, 16 Feb 2005
48 * - Advisory locking is ignored here.
49 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
53 #include <linux/module.h>
54 #include <linux/moduleparam.h>
55 #include <linux/sched.h>
57 #include <linux/file.h>
58 #include <linux/stat.h>
59 #include <linux/errno.h>
60 #include <linux/major.h>
61 #include <linux/wait.h>
62 #include <linux/blkdev.h>
63 #include <linux/blkpg.h>
64 #include <linux/init.h>
65 #include <linux/smp_lock.h>
66 #include <linux/swap.h>
67 #include <linux/slab.h>
68 #include <linux/loop.h>
69 #include <linux/compat.h>
70 #include <linux/suspend.h>
71 #include <linux/freezer.h>
72 #include <linux/writeback.h>
73 #include <linux/buffer_head.h> /* for invalidate_bdev() */
74 #include <linux/completion.h>
75 #include <linux/highmem.h>
76 #include <linux/gfp.h>
77 #include <linux/kthread.h>
78 #include <linux/splice.h>
80 #include <asm/uaccess.h>
82 static LIST_HEAD(loop_devices);
83 static DEFINE_MUTEX(loop_devices_mutex);
88 static int transfer_none(struct loop_device *lo, int cmd,
89 struct page *raw_page, unsigned raw_off,
90 struct page *loop_page, unsigned loop_off,
91 int size, sector_t real_block)
93 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
94 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
97 memcpy(loop_buf, raw_buf, size);
99 memcpy(raw_buf, loop_buf, size);
101 kunmap_atomic(raw_buf, KM_USER0);
102 kunmap_atomic(loop_buf, KM_USER1);
107 static int transfer_xor(struct loop_device *lo, int cmd,
108 struct page *raw_page, unsigned raw_off,
109 struct page *loop_page, unsigned loop_off,
110 int size, sector_t real_block)
112 char *raw_buf = kmap_atomic(raw_page, KM_USER0) + raw_off;
113 char *loop_buf = kmap_atomic(loop_page, KM_USER1) + loop_off;
114 char *in, *out, *key;
125 key = lo->lo_encrypt_key;
126 keysize = lo->lo_encrypt_key_size;
127 for (i = 0; i < size; i++)
128 *out++ = *in++ ^ key[(i & 511) % keysize];
130 kunmap_atomic(raw_buf, KM_USER0);
131 kunmap_atomic(loop_buf, KM_USER1);
136 static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
138 if (unlikely(info->lo_encrypt_key_size <= 0))
143 static struct loop_func_table none_funcs = {
144 .number = LO_CRYPT_NONE,
145 .transfer = transfer_none,
148 static struct loop_func_table xor_funcs = {
149 .number = LO_CRYPT_XOR,
150 .transfer = transfer_xor,
154 /* xfer_funcs[0] is special - its release function is never called */
155 static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
160 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
162 loff_t size, offset, loopsize;
164 /* Compute loopsize in bytes */
165 size = i_size_read(file->f_mapping->host);
166 offset = lo->lo_offset;
167 loopsize = size - offset;
168 if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
169 loopsize = lo->lo_sizelimit;
172 * Unfortunately, if we want to do I/O on the device,
173 * the number of 512-byte sectors has to fit into a sector_t.
175 return loopsize >> 9;
179 figure_loop_size(struct loop_device *lo)
181 loff_t size = get_loop_size(lo, lo->lo_backing_file);
182 sector_t x = (sector_t)size;
184 if (unlikely((loff_t)x != size))
187 set_capacity(lo->lo_disk, x);
192 lo_do_transfer(struct loop_device *lo, int cmd,
193 struct page *rpage, unsigned roffs,
194 struct page *lpage, unsigned loffs,
195 int size, sector_t rblock)
197 if (unlikely(!lo->transfer))
200 return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
204 * do_lo_send_aops - helper for writing data to a loop device
206 * This is the fast version for backing filesystems which implement the address
207 * space operations prepare_write and commit_write.
209 static int do_lo_send_aops(struct loop_device *lo, struct bio_vec *bvec,
210 int bsize, loff_t pos, struct page *page)
212 struct file *file = lo->lo_backing_file; /* kudos to NFsckingS */
213 struct address_space *mapping = file->f_mapping;
214 const struct address_space_operations *aops = mapping->a_ops;
216 unsigned offset, bv_offs;
219 mutex_lock(&mapping->host->i_mutex);
220 index = pos >> PAGE_CACHE_SHIFT;
221 offset = pos & ((pgoff_t)PAGE_CACHE_SIZE - 1);
222 bv_offs = bvec->bv_offset;
229 IV = ((sector_t)index << (PAGE_CACHE_SHIFT - 9))+(offset >> 9);
230 size = PAGE_CACHE_SIZE - offset;
233 page = grab_cache_page(mapping, index);
236 ret = aops->prepare_write(file, page, offset,
239 if (ret == AOP_TRUNCATED_PAGE) {
240 page_cache_release(page);
245 transfer_result = lo_do_transfer(lo, WRITE, page, offset,
246 bvec->bv_page, bv_offs, size, IV);
247 if (unlikely(transfer_result)) {
249 * The transfer failed, but we still write the data to
250 * keep prepare/commit calls balanced.
252 printk(KERN_ERR "loop: transfer error block %llu\n",
253 (unsigned long long)index);
254 zero_user_page(page, offset, size, KM_USER0);
256 flush_dcache_page(page);
257 ret = aops->commit_write(file, page, offset,
260 if (ret == AOP_TRUNCATED_PAGE) {
261 page_cache_release(page);
266 if (unlikely(transfer_result))
274 page_cache_release(page);
278 mutex_unlock(&mapping->host->i_mutex);
282 page_cache_release(page);
289 * __do_lo_send_write - helper for writing data to a loop device
291 * This helper just factors out common code between do_lo_send_direct_write()
292 * and do_lo_send_write().
294 static int __do_lo_send_write(struct file *file,
295 u8 *buf, const int len, loff_t pos)
298 mm_segment_t old_fs = get_fs();
301 bw = file->f_op->write(file, buf, len, &pos);
303 if (likely(bw == len))
305 printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
306 (unsigned long long)pos, len);
313 * do_lo_send_direct_write - helper for writing data to a loop device
315 * This is the fast, non-transforming version for backing filesystems which do
316 * not implement the address space operations prepare_write and commit_write.
317 * It uses the write file operation which should be present on all writeable
320 static int do_lo_send_direct_write(struct loop_device *lo,
321 struct bio_vec *bvec, int bsize, loff_t pos, struct page *page)
323 ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
324 kmap(bvec->bv_page) + bvec->bv_offset,
326 kunmap(bvec->bv_page);
332 * do_lo_send_write - helper for writing data to a loop device
334 * This is the slow, transforming version for filesystems which do not
335 * implement the address space operations prepare_write and commit_write. It
336 * uses the write file operation which should be present on all writeable
339 * Using fops->write is slower than using aops->{prepare,commit}_write in the
340 * transforming case because we need to double buffer the data as we cannot do
341 * the transformations in place as we do not have direct access to the
342 * destination pages of the backing file.
344 static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
345 int bsize, loff_t pos, struct page *page)
347 int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
348 bvec->bv_offset, bvec->bv_len, pos >> 9);
350 return __do_lo_send_write(lo->lo_backing_file,
351 page_address(page), bvec->bv_len,
353 printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
354 "length %i.\n", (unsigned long long)pos, bvec->bv_len);
360 static int lo_send(struct loop_device *lo, struct bio *bio, int bsize,
363 int (*do_lo_send)(struct loop_device *, struct bio_vec *, int, loff_t,
365 struct bio_vec *bvec;
366 struct page *page = NULL;
369 do_lo_send = do_lo_send_aops;
370 if (!(lo->lo_flags & LO_FLAGS_USE_AOPS)) {
371 do_lo_send = do_lo_send_direct_write;
372 if (lo->transfer != transfer_none) {
373 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
377 do_lo_send = do_lo_send_write;
380 bio_for_each_segment(bvec, bio, i) {
381 ret = do_lo_send(lo, bvec, bsize, pos, page);
393 printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
398 struct lo_read_data {
399 struct loop_device *lo;
406 lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
407 struct splice_desc *sd)
409 struct lo_read_data *p = sd->u.data;
410 struct loop_device *lo = p->lo;
411 struct page *page = buf->page;
416 ret = buf->ops->confirm(pipe, buf);
420 IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
426 if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
427 printk(KERN_ERR "loop: transfer error block %ld\n",
432 flush_dcache_page(p->page);
441 lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
443 return __splice_from_pipe(pipe, sd, lo_splice_actor);
447 do_lo_receive(struct loop_device *lo,
448 struct bio_vec *bvec, int bsize, loff_t pos)
450 struct lo_read_data cookie;
451 struct splice_desc sd;
456 cookie.page = bvec->bv_page;
457 cookie.offset = bvec->bv_offset;
458 cookie.bsize = bsize;
461 sd.total_len = bvec->bv_len;
466 file = lo->lo_backing_file;
467 retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
476 lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
478 struct bio_vec *bvec;
481 bio_for_each_segment(bvec, bio, i) {
482 ret = do_lo_receive(lo, bvec, bsize, pos);
490 static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
495 pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
496 if (bio_rw(bio) == WRITE)
497 ret = lo_send(lo, bio, lo->lo_blocksize, pos);
499 ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
504 * Add bio to back of pending list
506 static void loop_add_bio(struct loop_device *lo, struct bio *bio)
508 if (lo->lo_biotail) {
509 lo->lo_biotail->bi_next = bio;
510 lo->lo_biotail = bio;
512 lo->lo_bio = lo->lo_biotail = bio;
516 * Grab first pending buffer
518 static struct bio *loop_get_bio(struct loop_device *lo)
522 if ((bio = lo->lo_bio)) {
523 if (bio == lo->lo_biotail)
524 lo->lo_biotail = NULL;
525 lo->lo_bio = bio->bi_next;
532 static int loop_make_request(request_queue_t *q, struct bio *old_bio)
534 struct loop_device *lo = q->queuedata;
535 int rw = bio_rw(old_bio);
540 BUG_ON(!lo || (rw != READ && rw != WRITE));
542 spin_lock_irq(&lo->lo_lock);
543 if (lo->lo_state != Lo_bound)
545 if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
547 loop_add_bio(lo, old_bio);
548 wake_up(&lo->lo_event);
549 spin_unlock_irq(&lo->lo_lock);
553 spin_unlock_irq(&lo->lo_lock);
554 bio_io_error(old_bio, old_bio->bi_size);
559 * kick off io on the underlying address space
561 static void loop_unplug(request_queue_t *q)
563 struct loop_device *lo = q->queuedata;
565 clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags);
566 blk_run_address_space(lo->lo_backing_file->f_mapping);
569 struct switch_request {
571 struct completion wait;
574 static void do_loop_switch(struct loop_device *, struct switch_request *);
576 static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
578 if (unlikely(!bio->bi_bdev)) {
579 do_loop_switch(lo, bio->bi_private);
582 int ret = do_bio_filebacked(lo, bio);
583 bio_endio(bio, bio->bi_size, ret);
588 * worker thread that handles reads/writes to file backed loop devices,
589 * to avoid blocking in our make_request_fn. it also does loop decrypting
590 * on reads for block backed loop, as that is too heavy to do from
591 * b_end_io context where irqs may be disabled.
593 * Loop explanation: loop_clr_fd() sets lo_state to Lo_rundown before
594 * calling kthread_stop(). Therefore once kthread_should_stop() is
595 * true, make_request will not place any more requests. Therefore
596 * once kthread_should_stop() is true and lo_bio is NULL, we are
597 * done with the loop.
599 static int loop_thread(void *data)
601 struct loop_device *lo = data;
604 set_user_nice(current, -20);
606 while (!kthread_should_stop() || lo->lo_bio) {
608 wait_event_interruptible(lo->lo_event,
609 lo->lo_bio || kthread_should_stop());
613 spin_lock_irq(&lo->lo_lock);
614 bio = loop_get_bio(lo);
615 spin_unlock_irq(&lo->lo_lock);
618 loop_handle_bio(lo, bio);
625 * loop_switch performs the hard work of switching a backing store.
626 * First it needs to flush existing IO, it does this by sending a magic
627 * BIO down the pipe. The completion of this BIO does the actual switch.
629 static int loop_switch(struct loop_device *lo, struct file *file)
631 struct switch_request w;
632 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
635 init_completion(&w.wait);
637 bio->bi_private = &w;
639 loop_make_request(lo->lo_queue, bio);
640 wait_for_completion(&w.wait);
645 * Do the actual switch; called from the BIO completion routine
647 static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
649 struct file *file = p->file;
650 struct file *old_file = lo->lo_backing_file;
651 struct address_space *mapping = file->f_mapping;
653 mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
654 lo->lo_backing_file = file;
655 lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
656 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
657 lo->old_gfp_mask = mapping_gfp_mask(mapping);
658 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
664 * loop_change_fd switched the backing store of a loopback device to
665 * a new file. This is useful for operating system installers to free up
666 * the original file and in High Availability environments to switch to
667 * an alternative location for the content in case of server meltdown.
668 * This can only work if the loop device is used read-only, and if the
669 * new backing store is the same size and type as the old backing store.
671 static int loop_change_fd(struct loop_device *lo, struct file *lo_file,
672 struct block_device *bdev, unsigned int arg)
674 struct file *file, *old_file;
679 if (lo->lo_state != Lo_bound)
682 /* the loop device has to be read-only */
684 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
692 inode = file->f_mapping->host;
693 old_file = lo->lo_backing_file;
697 if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
700 /* new backing store needs to support loop (eg splice_read) */
701 if (!inode->i_fop->splice_read)
704 /* size of the new backing store needs to be the same */
705 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
709 error = loop_switch(lo, file);
722 static inline int is_loop_device(struct file *file)
724 struct inode *i = file->f_mapping->host;
726 return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
729 static int loop_set_fd(struct loop_device *lo, struct file *lo_file,
730 struct block_device *bdev, unsigned int arg)
732 struct file *file, *f;
734 struct address_space *mapping;
735 unsigned lo_blocksize;
740 /* This is safe, since we have a reference from open(). */
741 __module_get(THIS_MODULE);
749 if (lo->lo_state != Lo_unbound)
752 /* Avoid recursion */
754 while (is_loop_device(f)) {
755 struct loop_device *l;
757 if (f->f_mapping->host->i_rdev == lo_file->f_mapping->host->i_rdev)
760 l = f->f_mapping->host->i_bdev->bd_disk->private_data;
761 if (l->lo_state == Lo_unbound) {
765 f = l->lo_backing_file;
768 mapping = file->f_mapping;
769 inode = mapping->host;
771 if (!(file->f_mode & FMODE_WRITE))
772 lo_flags |= LO_FLAGS_READ_ONLY;
775 if (S_ISREG(inode->i_mode) || S_ISBLK(inode->i_mode)) {
776 const struct address_space_operations *aops = mapping->a_ops;
778 * If we can't read - sorry. If we only can't write - well,
779 * it's going to be read-only.
781 if (!file->f_op->splice_read)
783 if (aops->prepare_write && aops->commit_write)
784 lo_flags |= LO_FLAGS_USE_AOPS;
785 if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
786 lo_flags |= LO_FLAGS_READ_ONLY;
788 lo_blocksize = S_ISBLK(inode->i_mode) ?
789 inode->i_bdev->bd_block_size : PAGE_SIZE;
796 size = get_loop_size(lo, file);
798 if ((loff_t)(sector_t)size != size) {
803 if (!(lo_file->f_mode & FMODE_WRITE))
804 lo_flags |= LO_FLAGS_READ_ONLY;
806 set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
808 lo->lo_blocksize = lo_blocksize;
809 lo->lo_device = bdev;
810 lo->lo_flags = lo_flags;
811 lo->lo_backing_file = file;
812 lo->transfer = transfer_none;
814 lo->lo_sizelimit = 0;
815 lo->old_gfp_mask = mapping_gfp_mask(mapping);
816 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
818 lo->lo_bio = lo->lo_biotail = NULL;
821 * set queue make_request_fn, and add limits based on lower level
824 blk_queue_make_request(lo->lo_queue, loop_make_request);
825 lo->lo_queue->queuedata = lo;
826 lo->lo_queue->unplug_fn = loop_unplug;
828 set_capacity(lo->lo_disk, size);
829 bd_set_size(bdev, size << 9);
831 set_blocksize(bdev, lo_blocksize);
833 lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
835 if (IS_ERR(lo->lo_thread)) {
836 error = PTR_ERR(lo->lo_thread);
839 lo->lo_state = Lo_bound;
840 wake_up_process(lo->lo_thread);
844 lo->lo_thread = NULL;
845 lo->lo_device = NULL;
846 lo->lo_backing_file = NULL;
848 set_capacity(lo->lo_disk, 0);
849 invalidate_bdev(bdev);
850 bd_set_size(bdev, 0);
851 mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
852 lo->lo_state = Lo_unbound;
856 /* This is safe: open() is still holding a reference. */
857 module_put(THIS_MODULE);
862 loop_release_xfer(struct loop_device *lo)
865 struct loop_func_table *xfer = lo->lo_encryption;
869 err = xfer->release(lo);
871 lo->lo_encryption = NULL;
872 module_put(xfer->owner);
878 loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
879 const struct loop_info64 *i)
884 struct module *owner = xfer->owner;
886 if (!try_module_get(owner))
889 err = xfer->init(lo, i);
893 lo->lo_encryption = xfer;
898 static int loop_clr_fd(struct loop_device *lo, struct block_device *bdev)
900 struct file *filp = lo->lo_backing_file;
901 gfp_t gfp = lo->old_gfp_mask;
903 if (lo->lo_state != Lo_bound)
906 if (lo->lo_refcnt > 1) /* we needed one fd for the ioctl */
912 spin_lock_irq(&lo->lo_lock);
913 lo->lo_state = Lo_rundown;
914 spin_unlock_irq(&lo->lo_lock);
916 kthread_stop(lo->lo_thread);
918 lo->lo_backing_file = NULL;
920 loop_release_xfer(lo);
923 lo->lo_device = NULL;
924 lo->lo_encryption = NULL;
926 lo->lo_sizelimit = 0;
927 lo->lo_encrypt_key_size = 0;
929 lo->lo_thread = NULL;
930 memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
931 memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
932 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
933 invalidate_bdev(bdev);
934 set_capacity(lo->lo_disk, 0);
935 bd_set_size(bdev, 0);
936 mapping_set_gfp_mask(filp->f_mapping, gfp);
937 lo->lo_state = Lo_unbound;
939 /* This is safe: open() is still holding a reference. */
940 module_put(THIS_MODULE);
945 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
948 struct loop_func_table *xfer;
950 if (lo->lo_encrypt_key_size && lo->lo_key_owner != current->uid &&
951 !capable(CAP_SYS_ADMIN))
953 if (lo->lo_state != Lo_bound)
955 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
958 err = loop_release_xfer(lo);
962 if (info->lo_encrypt_type) {
963 unsigned int type = info->lo_encrypt_type;
965 if (type >= MAX_LO_CRYPT)
967 xfer = xfer_funcs[type];
973 err = loop_init_xfer(lo, xfer, info);
977 if (lo->lo_offset != info->lo_offset ||
978 lo->lo_sizelimit != info->lo_sizelimit) {
979 lo->lo_offset = info->lo_offset;
980 lo->lo_sizelimit = info->lo_sizelimit;
981 if (figure_loop_size(lo))
985 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
986 memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
987 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
988 lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
992 lo->transfer = xfer->transfer;
993 lo->ioctl = xfer->ioctl;
995 lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
996 lo->lo_init[0] = info->lo_init[0];
997 lo->lo_init[1] = info->lo_init[1];
998 if (info->lo_encrypt_key_size) {
999 memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1000 info->lo_encrypt_key_size);
1001 lo->lo_key_owner = current->uid;
1008 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1010 struct file *file = lo->lo_backing_file;
1014 if (lo->lo_state != Lo_bound)
1016 error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1019 memset(info, 0, sizeof(*info));
1020 info->lo_number = lo->lo_number;
1021 info->lo_device = huge_encode_dev(stat.dev);
1022 info->lo_inode = stat.ino;
1023 info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1024 info->lo_offset = lo->lo_offset;
1025 info->lo_sizelimit = lo->lo_sizelimit;
1026 info->lo_flags = lo->lo_flags;
1027 memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1028 memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1029 info->lo_encrypt_type =
1030 lo->lo_encryption ? lo->lo_encryption->number : 0;
1031 if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1032 info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1033 memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1034 lo->lo_encrypt_key_size);
1040 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1042 memset(info64, 0, sizeof(*info64));
1043 info64->lo_number = info->lo_number;
1044 info64->lo_device = info->lo_device;
1045 info64->lo_inode = info->lo_inode;
1046 info64->lo_rdevice = info->lo_rdevice;
1047 info64->lo_offset = info->lo_offset;
1048 info64->lo_sizelimit = 0;
1049 info64->lo_encrypt_type = info->lo_encrypt_type;
1050 info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1051 info64->lo_flags = info->lo_flags;
1052 info64->lo_init[0] = info->lo_init[0];
1053 info64->lo_init[1] = info->lo_init[1];
1054 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1055 memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1057 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1058 memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1062 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1064 memset(info, 0, sizeof(*info));
1065 info->lo_number = info64->lo_number;
1066 info->lo_device = info64->lo_device;
1067 info->lo_inode = info64->lo_inode;
1068 info->lo_rdevice = info64->lo_rdevice;
1069 info->lo_offset = info64->lo_offset;
1070 info->lo_encrypt_type = info64->lo_encrypt_type;
1071 info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1072 info->lo_flags = info64->lo_flags;
1073 info->lo_init[0] = info64->lo_init[0];
1074 info->lo_init[1] = info64->lo_init[1];
1075 if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1076 memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1078 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1079 memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1081 /* error in case values were truncated */
1082 if (info->lo_device != info64->lo_device ||
1083 info->lo_rdevice != info64->lo_rdevice ||
1084 info->lo_inode != info64->lo_inode ||
1085 info->lo_offset != info64->lo_offset)
1092 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1094 struct loop_info info;
1095 struct loop_info64 info64;
1097 if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1099 loop_info64_from_old(&info, &info64);
1100 return loop_set_status(lo, &info64);
1104 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1106 struct loop_info64 info64;
1108 if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1110 return loop_set_status(lo, &info64);
1114 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1115 struct loop_info info;
1116 struct loop_info64 info64;
1122 err = loop_get_status(lo, &info64);
1124 err = loop_info64_to_old(&info64, &info);
1125 if (!err && copy_to_user(arg, &info, sizeof(info)))
1132 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1133 struct loop_info64 info64;
1139 err = loop_get_status(lo, &info64);
1140 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1146 static int lo_ioctl(struct inode * inode, struct file * file,
1147 unsigned int cmd, unsigned long arg)
1149 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1152 mutex_lock(&lo->lo_ctl_mutex);
1155 err = loop_set_fd(lo, file, inode->i_bdev, arg);
1157 case LOOP_CHANGE_FD:
1158 err = loop_change_fd(lo, file, inode->i_bdev, arg);
1161 err = loop_clr_fd(lo, inode->i_bdev);
1163 case LOOP_SET_STATUS:
1164 err = loop_set_status_old(lo, (struct loop_info __user *) arg);
1166 case LOOP_GET_STATUS:
1167 err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1169 case LOOP_SET_STATUS64:
1170 err = loop_set_status64(lo, (struct loop_info64 __user *) arg);
1172 case LOOP_GET_STATUS64:
1173 err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1176 err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
1178 mutex_unlock(&lo->lo_ctl_mutex);
1182 #ifdef CONFIG_COMPAT
1183 struct compat_loop_info {
1184 compat_int_t lo_number; /* ioctl r/o */
1185 compat_dev_t lo_device; /* ioctl r/o */
1186 compat_ulong_t lo_inode; /* ioctl r/o */
1187 compat_dev_t lo_rdevice; /* ioctl r/o */
1188 compat_int_t lo_offset;
1189 compat_int_t lo_encrypt_type;
1190 compat_int_t lo_encrypt_key_size; /* ioctl w/o */
1191 compat_int_t lo_flags; /* ioctl r/o */
1192 char lo_name[LO_NAME_SIZE];
1193 unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1194 compat_ulong_t lo_init[2];
1199 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1200 * - noinlined to reduce stack space usage in main part of driver
1203 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1204 struct loop_info64 *info64)
1206 struct compat_loop_info info;
1208 if (copy_from_user(&info, arg, sizeof(info)))
1211 memset(info64, 0, sizeof(*info64));
1212 info64->lo_number = info.lo_number;
1213 info64->lo_device = info.lo_device;
1214 info64->lo_inode = info.lo_inode;
1215 info64->lo_rdevice = info.lo_rdevice;
1216 info64->lo_offset = info.lo_offset;
1217 info64->lo_sizelimit = 0;
1218 info64->lo_encrypt_type = info.lo_encrypt_type;
1219 info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1220 info64->lo_flags = info.lo_flags;
1221 info64->lo_init[0] = info.lo_init[0];
1222 info64->lo_init[1] = info.lo_init[1];
1223 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1224 memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1226 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1227 memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1232 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1233 * - noinlined to reduce stack space usage in main part of driver
1236 loop_info64_to_compat(const struct loop_info64 *info64,
1237 struct compat_loop_info __user *arg)
1239 struct compat_loop_info info;
1241 memset(&info, 0, sizeof(info));
1242 info.lo_number = info64->lo_number;
1243 info.lo_device = info64->lo_device;
1244 info.lo_inode = info64->lo_inode;
1245 info.lo_rdevice = info64->lo_rdevice;
1246 info.lo_offset = info64->lo_offset;
1247 info.lo_encrypt_type = info64->lo_encrypt_type;
1248 info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1249 info.lo_flags = info64->lo_flags;
1250 info.lo_init[0] = info64->lo_init[0];
1251 info.lo_init[1] = info64->lo_init[1];
1252 if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1253 memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1255 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1256 memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1258 /* error in case values were truncated */
1259 if (info.lo_device != info64->lo_device ||
1260 info.lo_rdevice != info64->lo_rdevice ||
1261 info.lo_inode != info64->lo_inode ||
1262 info.lo_offset != info64->lo_offset ||
1263 info.lo_init[0] != info64->lo_init[0] ||
1264 info.lo_init[1] != info64->lo_init[1])
1267 if (copy_to_user(arg, &info, sizeof(info)))
1273 loop_set_status_compat(struct loop_device *lo,
1274 const struct compat_loop_info __user *arg)
1276 struct loop_info64 info64;
1279 ret = loop_info64_from_compat(arg, &info64);
1282 return loop_set_status(lo, &info64);
1286 loop_get_status_compat(struct loop_device *lo,
1287 struct compat_loop_info __user *arg)
1289 struct loop_info64 info64;
1295 err = loop_get_status(lo, &info64);
1297 err = loop_info64_to_compat(&info64, arg);
1301 static long lo_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1303 struct inode *inode = file->f_path.dentry->d_inode;
1304 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1309 case LOOP_SET_STATUS:
1310 mutex_lock(&lo->lo_ctl_mutex);
1311 err = loop_set_status_compat(
1312 lo, (const struct compat_loop_info __user *) arg);
1313 mutex_unlock(&lo->lo_ctl_mutex);
1315 case LOOP_GET_STATUS:
1316 mutex_lock(&lo->lo_ctl_mutex);
1317 err = loop_get_status_compat(
1318 lo, (struct compat_loop_info __user *) arg);
1319 mutex_unlock(&lo->lo_ctl_mutex);
1322 case LOOP_GET_STATUS64:
1323 case LOOP_SET_STATUS64:
1324 arg = (unsigned long) compat_ptr(arg);
1326 case LOOP_CHANGE_FD:
1327 err = lo_ioctl(inode, file, cmd, arg);
1338 static int lo_open(struct inode *inode, struct file *file)
1340 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1342 mutex_lock(&lo->lo_ctl_mutex);
1344 mutex_unlock(&lo->lo_ctl_mutex);
1349 static int lo_release(struct inode *inode, struct file *file)
1351 struct loop_device *lo = inode->i_bdev->bd_disk->private_data;
1353 mutex_lock(&lo->lo_ctl_mutex);
1355 mutex_unlock(&lo->lo_ctl_mutex);
1360 static struct block_device_operations lo_fops = {
1361 .owner = THIS_MODULE,
1363 .release = lo_release,
1365 #ifdef CONFIG_COMPAT
1366 .compat_ioctl = lo_compat_ioctl,
1371 * And now the modules code and kernel interface.
1373 static int max_loop;
1374 module_param(max_loop, int, 0);
1375 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1376 MODULE_LICENSE("GPL");
1377 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1379 int loop_register_transfer(struct loop_func_table *funcs)
1381 unsigned int n = funcs->number;
1383 if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1385 xfer_funcs[n] = funcs;
1389 int loop_unregister_transfer(int number)
1391 unsigned int n = number;
1392 struct loop_device *lo;
1393 struct loop_func_table *xfer;
1395 if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1398 xfer_funcs[n] = NULL;
1400 list_for_each_entry(lo, &loop_devices, lo_list) {
1401 mutex_lock(&lo->lo_ctl_mutex);
1403 if (lo->lo_encryption == xfer)
1404 loop_release_xfer(lo);
1406 mutex_unlock(&lo->lo_ctl_mutex);
1412 EXPORT_SYMBOL(loop_register_transfer);
1413 EXPORT_SYMBOL(loop_unregister_transfer);
1415 static struct loop_device *loop_alloc(int i)
1417 struct loop_device *lo;
1418 struct gendisk *disk;
1420 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1424 lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1428 disk = lo->lo_disk = alloc_disk(1);
1430 goto out_free_queue;
1432 mutex_init(&lo->lo_ctl_mutex);
1434 lo->lo_thread = NULL;
1435 init_waitqueue_head(&lo->lo_event);
1436 spin_lock_init(&lo->lo_lock);
1437 disk->major = LOOP_MAJOR;
1438 disk->first_minor = i;
1439 disk->fops = &lo_fops;
1440 disk->private_data = lo;
1441 disk->queue = lo->lo_queue;
1442 sprintf(disk->disk_name, "loop%d", i);
1446 blk_cleanup_queue(lo->lo_queue);
1453 static void loop_free(struct loop_device *lo)
1455 blk_cleanup_queue(lo->lo_queue);
1456 put_disk(lo->lo_disk);
1457 list_del(&lo->lo_list);
1461 static struct loop_device *loop_init_one(int i)
1463 struct loop_device *lo;
1465 list_for_each_entry(lo, &loop_devices, lo_list) {
1466 if (lo->lo_number == i)
1472 add_disk(lo->lo_disk);
1473 list_add_tail(&lo->lo_list, &loop_devices);
1478 static void loop_del_one(struct loop_device *lo)
1480 del_gendisk(lo->lo_disk);
1484 static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1486 struct loop_device *lo;
1487 struct kobject *kobj;
1489 mutex_lock(&loop_devices_mutex);
1490 lo = loop_init_one(dev & MINORMASK);
1491 kobj = lo ? get_disk(lo->lo_disk) : ERR_PTR(-ENOMEM);
1492 mutex_unlock(&loop_devices_mutex);
1498 static int __init loop_init(void)
1501 unsigned long range;
1502 struct loop_device *lo, *next;
1505 * loop module now has a feature to instantiate underlying device
1506 * structure on-demand, provided that there is an access dev node.
1507 * However, this will not work well with user space tool that doesn't
1508 * know about such "feature". In order to not break any existing
1509 * tool, we do the following:
1511 * (1) if max_loop is specified, create that many upfront, and this
1512 * also becomes a hard limit.
1513 * (2) if max_loop is not specified, create 8 loop device on module
1514 * load, user can further extend loop device by create dev node
1515 * themselves and have kernel automatically instantiate actual
1518 if (max_loop > 1UL << MINORBITS)
1526 range = 1UL << MINORBITS;
1529 if (register_blkdev(LOOP_MAJOR, "loop"))
1532 for (i = 0; i < nr; i++) {
1536 list_add_tail(&lo->lo_list, &loop_devices);
1539 /* point of no return */
1541 list_for_each_entry(lo, &loop_devices, lo_list)
1542 add_disk(lo->lo_disk);
1544 blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1545 THIS_MODULE, loop_probe, NULL, NULL);
1547 printk(KERN_INFO "loop: module loaded\n");
1551 printk(KERN_INFO "loop: out of memory\n");
1553 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1556 unregister_blkdev(LOOP_MAJOR, "loop");
1560 static void __exit loop_exit(void)
1562 unsigned long range;
1563 struct loop_device *lo, *next;
1565 range = max_loop ? max_loop : 1UL << MINORBITS;
1567 list_for_each_entry_safe(lo, next, &loop_devices, lo_list)
1570 blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1571 unregister_blkdev(LOOP_MAJOR, "loop");
1574 module_init(loop_init);
1575 module_exit(loop_exit);
1578 static int __init max_loop_setup(char *str)
1580 max_loop = simple_strtol(str, NULL, 0);
1584 __setup("max_loop=", max_loop_setup);